Delivery case: AI inside an offline company network

A Shenzhen client needed a large model to run in its own data centre without internet access. JSLE deployed GLM-5.2-FP8 across two servers with 16 H200 GPUs and provided a shared connection for business tools. Basic functional checks and project acceptance were completed.

Successful delivery case · Deployment, functional verification and acceptance complete

GPU scale2 inference nodes · 16 H200 GPUs

Target modelGLM-5.2-FP8

Inference stackRay + vLLM across two nodes

EnvironmentCustomer network · Offline implementation

Project facts

Background and boundaries

The customer wanted controlled inference in its own data centre, with model weights, invocation paths and runtime logs kept in the enterprise network. Client name, domains, people, server addresses and security settings are redacted; this page presents only the anonymised architecture, delivery method and test conclusions.

Functional verification

All six basic functional tests passed. Four concurrent SSE streams were verified and independent sessions remained isolated. Performance benchmarks, long-context behaviour and higher concurrency remain subject to agreed test conditions and formal acceptance evidence.

Relationship to law firm work

The offline deployment, unified OpenAI-compatible interface and permission-controlled workspace (Hermes Agent with multi-user workspaces) delivered in this case apply directly to law firm document search and access control: the model and the material stay inside the firm's own network, and each user works within an agreed boundary.

Matter and contract ingestion, per-matter and per-user permission mapping, retrieval quality acceptance and answer traceability are incremental work on top of this foundation. JSLE has not yet completed an accepted law firm engagement, so this case evidences deployment and access-control capability rather than an industry track record.

Next step

For a multi-node planning discussion, we can provide capacity sizing, implementation and acceptance planning.

Phone: +86 139 2521 1225 Email: tianjun@jsle.cn